Sequence of labels about fold

[delesley]

If I understand you correctly, you're asking how to add a loss function to the RNN such that the output of the RNN is used to predict the labels, correct? You can use td.Metric to record a loss for each item, and compiler.metric_tensors to read the loss tensors back out and feed them to the optimizer.

…

On Fri, Feb 24, 2017 at 11:26 AM, William Guss ***@***.***> wrote: Hi suppose I have an RNN where at each time step I have labels for the output, in this case how might I use fold to set up the computation graph. I am aware that we can get the last output state of the RNN using td.RNN, but is there a way to accomplish what I described? — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#18>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AGGbTUYj-t9PIdiweS16JyZOXRsr0Km-ks5rfy7zgaJpZM4MLlWQ> .

-- DeLesley Hutchins | Software Engineer | [email protected] | 505-206-0315

[MadcowD]

Do you have a full example of its use with an RNN with labels at every timestep? The documentation is helpful, but still doesn't provide an entire picture for me.

Thanks :)

[noa]

You might find the sentiment example useful since it shows how to use td.Metric to compute a loss for the recursive case. The sequence case is the same except you don't need to worry about recursion and just apply the metric at every position in the sequence.

[MadcowD]

Thanks!